Previous studies reported a good performance of video oculography compared with scleral search coils. Van der Geest and Frens
12 compared the performance of a 2D video-based eye tracker (Eyelink I; SR Research Ltd., Mississauga, Ontario, Canada) with 2D scleral search coils. They found a very good correspondence between the video and the coil output, with a high correlation of fixation positions (average discrepancy, <1° over a tested range of 40 by 40° of visual angle) and linear fits near one (range, 0.994 to 1.096) for saccadic properties. However, the video system they used, with a sampling rate of 250 Hz, was not capable of measuring torsion. Clarke et al.
13 tested the Chronos system we used. They did this in a highly conditional setup comprising an artificial eye in front of an imaging camera. The artificial eye, with clear iris landmarks, was mounted on a three-axis gimbal and aligned orthogonally to the imaging camera. The measurement resolution for the horizontal, vertical, and torsional positions was 0.006°, 0.005°, and 0.016°, respectively. In this ideal setup, measurement error for positions in the range of −20° to 20° was 0.1° for horizontal and vertical positions and 0.4° for torsional positions. They also compared the Chronos system with scleral search coils, much like our verification test with fixations. Subjects successively fixated on targets arranged in a 5° interval grid with horizontal range of −20° to 20° and vertical range of −15° to 15°. Eye movements were recorded simultaneously by scleral search coils and by the Chronos system. In the coils that they used, black markers were embedded. Offline, eye positions were calculated by tracking these markers with marker tracking software. The eye positions were recorded at a sampling rate of 50 Hz instead of 200 Hz. The measured system noise was in the order of 0.1° for both coil and video signals. This is in contrast to the smaller noise levels of coils compared with video signals we found. One problem when comparing signal noises in the two systems is that the sources of noise are different. In coil signals, there is a physically extremely low signal noise, which is dependent on the magnetic field strength and the number of coil windings. The same applies to the extraction of a position signal with the Chronos video system from a stationary artificial eye. The differences arise when one wants to measure the movements of a real eye in human subjects. Variations in coil signals are mainly determined by the ability of the subjects to hold fixation. This means that the signal noise of a coil attached to a real eye calculated over a short time span is low (<0.02°). In contrast, because the extraction of position signals by the video system is based on tracking of the pupil with continuous variations in diameter, video signals show a larger variability, which is inherent to measuring a biological signal. A comparison of noise levels over longer periods of time gives another picture, because then the noise is more governed by fixation stability. When we calculated the noise by taking the SD around the mean during each fixation interval and averaging over all targets and all eyes, average noise levels for coil signals during fixation were 0.25°, 0.10°, and 0.22° for horizontal, vertical, and torsional eye positions. For the Chronos signals measured at 200 Hz, these values were 0.32°, 0.26°, and 0.41°. These numbers are higher than the value of 0.1° reported by Clarke et al.,
13 for both coil and video signals, which may be partially due to the lower sampling rate of the video signals in their experiment.